Using Artificial Neural Networks to Predict Intra-Abdominal Abscess Risk Post-Appendectomy

Area of Science:

• Surgical outcomes research
• Artificial intelligence in medicine
• Pediatric surgery

Background:

• Intra-abdominal abscess (IAA) affects 13.6%–14.6% of appendicitis cases, often linked to complicated appendicitis.
• Inconsistent appendicitis severity classification and treatment variations, particularly for perforated appendicitis, highlight a need for improved risk stratification.

Purpose of the Study:

• To evaluate the efficacy of artificial neural networks (ANN) in predicting the risk of intra-abdominal abscess (IAA) development post-appendectomy.
• To develop reproducible and explainable ANN models for enhanced clinical decision-making in pediatric appendectomy patients.

Main Methods:

• Developed two distinct artificial neural network (ANN) architectures using retrospective data from 1574 pediatric patients (<19 years old) who underwent appendectomy.
• Utilized demographic, clinical, and surgical variables, selecting the 12 most influential factors for model training and testing.
• Employed an 80%/20% split for training and testing datasets to validate model performance.

Main Results:

• Model 1 demonstrated high accuracy (89.84%), with 70% sensitivity and 93.61% specificity in predicting IAA on the test set.
• Model 2 achieved 84.13% accuracy, 81.63% sensitivity, and 84.6% specificity, indicating robust predictive capabilities.
• Both ANN models showed significant potential in identifying patients at risk for post-appendectomy IAA.

Conclusions:

• Artificial neural networks (ANN), when applied to carefully selected variables, can accurately predict the likelihood of intra-abdominal abscess (IAA) following appendectomy.
• The developed ANN models are reproducible and explainable, offering a potential state-of-the-art approach for optimizing post-appendectomy care and patient management.